Machine learning for localization in LoRaWAN: a case study with data augmentation

Abstract

The growth of Internet of Things applications such as smart cities, leads to an increase in the number of connected objects. In some cases, a requirement of such applications is the location of devices for monitoring and management. This paper develops a methodology for the location of different nodes based on the signal levels received in a LoRaWAN network. The goal is to detect changes in node positions of at least 100 m with a limited amount of data. The procedure involves data analysis, preprocessing, and evaluation of different machine learning algorithms to locate the nodes. Due to the large data volume requirements for the selected algorithms, the work includes the application of a simple-to-implement data augmentation technique. As a result, the best performing algorithm was K Nearest Neighbors with an average error of 12 m.